Americans Have Turned Against AI — Transcript

Full transcript

1. 0:00Millions of Americans will open  ChatGPT in the next 10 minutes…
2. 0:03And most of them believe it's making  the world worse. Half the country uses
3. 0:07AI now. Yet only 16% of Americans  think it will be good for society.
4. 0:12It’s the lowest level of optimism any  major technology has launched with.
5. 0:16The people building AI call it humanity's greatest
6. 0:18breakthrough. The people using it describe  something closer to a hostage situation.
7. 0:23So how did the fastest-adopted technology of  our lifetime also become one of the most hated?
8. 0:28You might expect that it’s the older  generations that are pushing back on
9. 0:31the AI revolution. After all, every new  technology has faced resistance from
10. 0:36people who didn't grow up with it. But that's not what's happening.
11. 0:39The biggest users of AI are Americans  under 30. They use it for schoolwork,
12. 0:44job applications, coding, brainstorming,  and as a substitute for Google. By every
13. 0:49indicator, they should be AI's biggest fans. Instead, they're some of its biggest critics.
14. 0:54A Gallup study found the share of Gen Z who  say AI makes them feel angry climbed from
15. 0:5822% to 31% by early 2026. And according to Pew  Research, 40% of young Americans said it would
16. 1:05make the country worse than better. But this isn't just a Gen Z story.
17. 1:09Trust in AI is collapsing all  across the U.S. Two thirds of
18. 1:13people think it's being developed too  quickly. 59% don't trust the companies
19. 1:17building it. And most don't believe the  government can keep it under control.
20. 1:20Yet people keep using it. ChatGPT alone is  used by 44% of American adults, more than
21. 1:27double its share in 2023. That's the AI paradox.
22. 1:31So why does adoption keep accelerating  if confidence keeps falling?
23. 1:34For most people, it wasn’t really adopted.  Workplaces began implementing it in their
24. 1:39workflow. Schools added it to assignments. Apps  and search engines displayed it as a main feature.
25. 1:44People had no other choice. It was forced  upon them. AI didn’t earn their trust.
26. 1:49They have no other option.
27. 1:51The shift happened so gradually  that most people barely noticed it.
28. 1:54Until suddenly, it was everywhere.
29. 1:56The moment you open Google and search  for something, you’re hit with a block
30. 2:00of AI-written text answering the question  before you even reach the search results.
31. 2:04The company that’s supposed to help you find  information is now deciding what you see first.
32. 2:09And it works.
33. 2:1060% of Americans say they read the summaries  regularly, because there’s no way to avoid
34. 2:15them. The moment Google started answering  the question itself, the experience changed.
35. 2:20And the more the internet degraded, the less  people trusted it. In 2021, 37% of Americans said
36. 2:26the spread of AI worried them more than it excited  them. By 2025, that number had climbed to 50%,
37. 2:32even though most people hadn't been personally  affected by AI in a professional capacity.
38. 2:37That would soon change.
39. 2:39In April 2025, Tobi Lütke,  Shopify’s chief executive,
40. 2:43posted a memo to his staff. Before any team could  ask him for a bigger budget or another employee,
41. 2:47they had to prove that a machine  couldn’t do the work first. Lütke
42. 2:51insisted that “reflexive AI use  was the new baseline expectation”.
43. 2:55Shopify wasn’t the only  company going all in on AI.
44. 2:59Language learning platform Duolingo was next. CEO,  Luis von Ahn, told his staff the company was going
45. 3:04“AI-first.” It would rely on AI before hiring,  reduce the number of contractors, and weigh AI
46. 3:10use in performance reviews. Within a week, 148 new  courses were shipped, it was work that would take
47. 3:17staff years to develop. The same AI implementation  turned up at Box, at Fiverr, and other firms.
48. 3:23The wording barely changed. Run the software  before you ask for new hires or extra money.
49. 3:28Then the pushback happened.
50. 3:29Longtime Duolingo users filmed themselves deleting  the app in protest. Staff started asking if they
51. 3:34had to use AI well or just use it. Almost a year  later Luis von Ahn walked his decision back.
52. 3:40He acknowledged that the AI-written code was  hard to debug and the lessons weren’t reliable.
53. 3:45But the damage had been done.
54. 3:47And it was just the beginning.
55. 3:48The fear of AI taking jobs is real, so in the  summer of 2025, a Stanford economist named Erik
56. 3:54Brynjolfsson decided to look into it. His team at  the Digital Economy Lab pulled the records from
57. 3:59ADP, the firm that processes the payroll for  millions of American workers, and studied who
58. 4:04still had a job, month after month. The AI impact was real.
59. 4:09Just now where people expected it.
60. 4:11The results showed that workers aged 22 to 25  in software and customer support had dropped
61. 4:16about 13% since late 2022. Workers over  30 in the same roles remained stable or
62. 4:22actually grew. It was the people just walking  in the door that were at risk of an AI cull.
63. 4:27Some don’t get in the door at all.
64. 4:29Revelio Labs, which tracks hiring across the  economy, counts roughly 35% fewer entry-level
65. 4:35postings since the start of 2023. Once  employers saw what the best models could do,
66. 4:40many stopped graduate hiring. That freeze only  surfaced in the payroll data a year or 2 later.
67. 4:46Brynjolfsson’s team found that when a company  used AI to take over a task completely,
68. 4:51employment among younger workers fell.  When AI was used to make a person faster,
69. 4:55employment remained stable, and sometimes grew.
70. 4:58Same technology, opposite result.
71. 5:00Everything comes down to one decision:
72. 5:02does a company use AI to replace  a worker, or to back one up?
73. 5:05Near the bottom of the pay  scale, replacement is winning.
74. 5:08Salesforce froze much of its junior  hiring for a stretch and credited its
75. 5:12own AI for work graduates used  to do. Goldman Sachs economists
76. 5:16found the results. The big names  weren’t even pretending anymore.
77. 5:19But not every bet paid off.
78. 5:21Klarna, the buy-now-pay-later company, spent a  year bragging that its chatbot did the work of
79. 5:26hundreds of agents and that it had stopped hiring  people. They later admitted the bot gave “lower
80. 5:32quality” service and started hiring humans back.  But some companies didn’t see that as a reason
81. 5:36to slow down. They saw it as a reason to aim  lower… and the contractor was in the cross hairs.
82. 5:42An MIT report on AI in business found that AI  is wiping out outsourced and contract roles,
83. 5:47not full-time staff. In mid-2025, the  data-labeling firm Scale AI laid off 200
84. 5:53staff and cut ties with 500 contractors.  That same report put the share of jobs
85. 5:57AI can already do at roughly 3% today,  and saw it climbing over the longer run.
86. 6:02Graduates weren’t losing out  to people with a better resume.
87. 6:05They were losing out period. There was not job.
88. 6:09By early 2026, nearly half of employed Gen Z said  AI's risks at work now outweigh its benefits,
89. 6:14a jump of 11 percentage points in just one year.  And when they're handed the same piece of work,
90. 6:2069% overwhelmingly trust the human version over  the AI one. An entire generation now spends its
91. 6:26days training the technology that's replacing  the jobs they're trying to build careers in.
92. 6:31The lost jobs are easy enough to measure. But the resentment isn't just about money.
93. 6:35In July 2024, all it took was a power surge  across northern Virginia for nearly 60 data
94. 6:40centers to drop off the grid. Roughly 1,500  megawatts of demand vanished instantly.
95. 6:45Operators scrambled to get the surge under  control before it took down anything else.
96. 6:49For a few seconds, the densest data-center  hub in the country just wasn’t there.
97. 6:54Virginia holds more of these  buildings than anywhere else,
98. 6:57and by 2023 they were already drawing about  26% of the state’s electricity. That’s why
99. 7:02one voltage swing could shock the whole grid. Each of those buildings runs hot. One question
100. 7:08to a chatbot burns about 2.9 watt-hours,  that’s roughly 10 times a normal web search.
101. 7:13On its own, it’s nothing. But nobody asks just one question.
102. 7:17Multiply it by billions of questions a day, every  day, in warehouses that never switch off, and it
103. 7:22becomes 183 terawatt-hours a year. That’s about  4% of all the power the country burns. That's
104. 7:28already 540 kilowatt-hours running in your name  every year, whether you've ever touched a chatbot
105. 7:34or not. And the forecasts are only going up, as  high as 17% of the nation's electricity by 2030.
106. 7:41More than 40% of the power fuelling the data  centres comes from natural gas. Before the decade
107. 7:46is out, these buildings are set to burn more  electricity than the entire country uses to make
108. 7:50its steel, cement, aluminum, and chemicals… Combined.
109. 7:54A high-voltage line takes up to 8 years to build,  meaning a fifth of these data center projects are
110. 7:59about to run straight into a power shortage.  To cover the expense of these extra lines,
111. 8:04utilities are going to regulators to demand  more money from regular consumers. This means
112. 8:09your monthly expenses are going up by 2030 just to  keep the lights on at a server farm down the road.
113. 8:15You didn't leave your lights on or  crank up your AC, but your money
114. 8:18is bleeding out anyway to power the  massive cooling fans of a tech giant.
115. 8:22Researchers at Carnegie Mellon estimate  it could push the average American
116. 8:26power bill up about 8% by 2030, and in the  worst-hit parts of Virginia, north of 25%.
117. 8:33There simply isn’t enough  energy or money to fund it.
118. 8:36So, some companies got tired of waiting.
119. 8:38Instead of waiting years for the power  grid to catch up, they built their own
120. 8:42power plants beside the data centers. And  one company took that idea to the extreme.
121. 8:47To power Colossus, the supercomputer  training Elon Musk's Grok chatbot,
122. 8:51xAI deployed a fleet of gas turbines in  South Memphis.The entire site was up and
123. 8:56running in under 122 days, a project that  would normally take years. By the time the
124. 9:01Southern Environmental Law Center flew thermal  drones overhead, it counted around 35 turbines,
125. 9:06most of them without air permits. Just downwind sits Boxtown,
126. 9:10a neighborhood that has spent decades  surrounded by refineries and chemical plants.
127. 9:14Residents began showing up at public hearings  describing a rotten-egg smell in the air and
128. 9:19children wheezing through the night. Memphis  already earns an "F" for ozone pollution from
129. 9:24the American Lung Association, and Boxtown's  cancer risk is about 4 times the national average.
130. 9:29Then researchers at the University of Tennessee  analyzed satellite data for TIME and found peak
131. 9:34nitrogen dioxide levels near the site had jumped  roughly 79% compared with before xAI arrived.
132. 9:41The turbines were labelled as temporary,
133. 9:43a workaround that allowed them to keep running  and bypass the regulations. Lawsuits followed,
134. 9:48but by the time the case moved forward, the  servers had already been running for a year.
135. 9:52Power isn't the only resource  these data centers consume.
136. 9:55In Newton County, Georgia, about  an hour and a half east of Atlanta,
137. 9:59Beverly and Jeff Morris live 1,000 feet (305  meters) from a Meta data center. Their house
138. 10:04runs on a private well, which means their  drinking water comes straight out of the
139. 10:07ground. It worked fine for years, right up  until Meta broke ground next door in 2018.
140. 10:13A few months after construction began,  the water pressure slowed to a trickle,
141. 10:17then stopped altogether. When it eventually  came back, it was brown and murky. Beverly
142. 10:22Morris told the New York Times she was afraid  to drink the water coming from her own tap.
143. 10:26The neighboring data center uses roughly  500,000 gallons every day, nearly a tenth
144. 10:31of all the water Newton County pumps. County  projections show demand outpacing supply by 2030,
145. 10:37while residents face water bills  expected to jump 33% in just 2 years,
146. 10:41compared with the usual 2% increase. A Meta  funded study said the company’s operations
147. 10:46probably aren't to blame and has invested more  than $4.5 million in local schools and nonprofits.
148. 10:52Private wells like the Morrises' fall  outside the federal Safe Drinking Water Act,
149. 10:57so Meta didn’t break any laws. The Environmental  Protection Agency is powerless, so Meta keeps
150. 11:02pulling water to cool the chips. And these hubs are insatiable.
151. 11:07A 2023 study by UC Riverside and the University  of Texas found that training one model, GPT-3, ran
152. 11:13through about 5.4 million liters of fresh water…  before the public ever typed a word. The same
153. 11:20researchers found that a short back-and-forth,  around 100 words, consumes close to a full bottle.
154. 11:25One Google data center in Papillion, Nebraska,  went through about 1.5 billion liters in 2024,
155. 11:31more than 600 Olympic pools, and it’s one site  among thousands. That daily draw dwarfs the
156. 11:37one-time cost of training. Nationally, data  centers used over 60 billion liters in 2023,
157. 11:42and the World Resources Institute expects AI  to pass a trillion gallons a year by 2030.
158. 11:47In August 2025, Tucson’s city council  voted 7 to 0 to reject Project Blue,
159. 11:53an Amazon-linked data-center campus  that would have drained millions of
160. 11:56gallons of city water in the middle of a  desert. Residents had packed the chamber
161. 12:01for weeks under a banner that read  No Desert Data Center. For one night,
162. 12:05it looked like a regular city could tell a  tech giant no, and actually make it stick.
163. 12:10It didn’t end there.
164. 12:11The developer already owned land just outside  the city limits and kept building anyway. Then,
165. 12:16in the spring of 2026, the contractor was  caught drawing enough Tucson water to supply
166. 12:215 households for a year. The city shut it  off. Mayor, Regina Romero, has been writing
167. 12:26ordinances to guard against the next data  center, because there’s always a next one.
168. 12:31Tucson wasn’t alone.
169. 12:32In The Dalles, Oregon, a Google plant ended up  using close to a quarter of the entire town’s
170. 12:37water. The company fought to keep those numbers  hidden from the public through court orders.
171. 12:41Newer facilities can cool with air or  closed-loop systems that use almost no
172. 12:46local water. Microsoft has already converted  2 Arizona sites to do exactly that. But the
173. 12:51biggest water users aren't the new builds,  they're the older data centers. The ones
174. 12:55still sucking up millions of gallons  instead of returning it to the river.
175. 12:58So while families shorten their showers  and wait until dark to water the lawn,
176. 13:02the data center down the road keeps pulling  water at full capacity, every hour, every night.
177. 13:07Jobs. Electricity. Water. Three separate  battles happening in different parts of
178. 13:11the country. But they all lead back  to the exact same business model.
179. 13:14In 2024 alone, Amazon, Microsoft, Google, and Meta  spent more than $200 billion building the data
180. 13:20centers behind AI, up over 60% in a single year.  Management consultancy firm McKinsey expects $5.2
181. 13:27trillion to be funnelled into them by 2030. That  money buys the land, the power, and the water.
182. 13:33But there’s one more thing it buys. Something  nobody signs off on: the writing, art,
183. 13:38and code scraped off the open internet  to train the models in the first place.
184. 13:42To the companies, that’s an investment  in the platform of the century. To the
185. 13:46creatives whose work fed it, it  looks different. Their work was
186. 13:50harvested for free, then rented  back to everyone by the month.
187. 13:54Not one of them was asked.  Not one of them got paid.
188. 13:57Publishers and authors have sued in waves arguing  their work was stolen and used to train AI without
189. 14:02permission. The courts still haven't decided  where the legal line is, and so far none of
190. 14:06it has been ruled illegal. But AI is built on  other people's writing, powered by other people's
191. 14:12electricity, cooled with other people's water… Then sold back to those same people as progress.
192. 14:17Now, Americans are starting to notice. The  companies insist the costs are temporary.
193. 14:21Google and Microsoft both promise they'll return  more water than they use by 2030 and keep shifting
194. 14:27to cleaner energy. But if you've already lost  your job, your power bill just went up, or your
195. 14:32town's water is running low, a promise four years  from now doesn't feel like much of an answer.
196. 14:36Between May 2024 and March 2025, that anger  blocked or delayed about $64 billion in
197. 14:43data-center projects nationwide, according  to Data Center Watch. In Virginia alone,
198. 14:4842 separate groups were working  to slow or stop new ones.
199. 14:51What nobody agrees on is what to do about it.
200. 14:54No one is demanding an outright  ban. No one is cheering it on.
201. 14:58There’s just a country caught between  needing the tool and resenting it.
202. 15:02Most of these fights end the same  way. A council denies the water,
203. 15:05the way Tucson did. The county next door sells  the land. A state commission signs off on the
204. 15:11power. The building goes up anyway, trading its  thirst for water for a heavier pull on the grid.
205. 15:16Residents keep learning the same lesson: the  project can almost always route around them.
206. 15:20This was never about whether AI would  arrive.That was inevitable. It was about
207. 15:24whether anyone still had the power to stop it. By the time we started asking that question,
208. 15:29the answer was already no. It’s not just ordinary Americans
209. 15:32who are losing faith in AI. The people building  it are too. Watch Why AI Researchers Are Quitting
210. 15:38and Panicking on the Way Out to find out why the  industry is collapsing. Or click on this video.